Two Kinds of PDF That Look the Same
Open two PDFs side by side and they may look identical. Try to select a sentence with the mouse, and the difference appears. In one, the text highlights letter by letter. In the other, nothing happens, or the whole page is selected as one block.
The first is a text PDF, made by exporting from Word, a browser or any other program. It contains the actual characters together with their positions and fonts. The second is a scanned PDF: each page is a photograph of paper. To a computer, that page is a picture of a document in the same way a holiday photo is a picture of a beach. There are no letters in it, only pixels.
A quick test: press Ctrl+F and search for a word you can see on the page. If nothing is found, you have a scan.
Why the Difference Matters for Conversion
Converting a text PDF to Word is a layout problem. The characters are there; the converter has to work out which ones form a paragraph, where a table starts, and what was a heading. That works well for letters, reports and contracts.
With a scan there is nothing to lay out. A converter that does not recognise this gives you a Word file with one large picture per page — it opens, and it is exactly as uneditable as before. To get real text, the pictures have to be read first. That step is called OCR, optical character recognition.
What OCR Actually Does
OCR software works through a page in stages. It straightens the image and separates dark ink from the light background. It finds blocks of text, then lines, then words. It compares the shapes of the letters with what it has learned, and finally uses a dictionary of the chosen language to settle doubtful cases — deciding, for example, that "rn" next to each other is more likely an "m" in one word and really "rn" in another.
The language setting matters more than people expect. Recognition that only knows English will turn š into s or ü into u, and will "correct" foreign words into English ones. On this site, pages without a text layer are read with English, German, French, Italian, Spanish, Bosnian, Croatian, Serbian, Russian and Turkish loaded together.
What You Get Back
When you put a scanned PDF through PDF to Word here, the converter first checks whether the file has a text layer. If it does, the layout is rebuilt directly. If it does not, the pages go through OCR and you receive the recognised text as an editable document, page by page.
It is worth being clear about what that means. You get the words: paragraphs you can edit, search, copy and correct. You do not get a replica of the page design. Logos, stamps, signatures, letterheads and the exact position of every box are not rebuilt from a scan. For most purposes — quoting a contract, reusing a text, filling an archive with searchable content — the words are what is needed.
If you only need the text and no formatting at all, a plain text file is the lighter result; note that PDF to Text extracts the text layer, so it suits text PDFs rather than scans.
What Decides the Accuracy
| Factor | Good | Causes errors |
|---|---|---|
| Resolution | 300 dpi | Below 150 dpi; phone photos taken from far away |
| Sharpness and contrast | Black text on white paper | Blur, shadows, faded or grey print, coloured backgrounds |
| Alignment | Page lies straight | Tilted pages, curved book pages near the spine |
| Typeface | Ordinary printed fonts, 10 pt and larger | Very small print, decorative fonts, typewriter with worn ribbon |
| Content | Running text | Handwriting, formulas, dense tables, stamps over text |
A clean office scan of a printed letter is typically recognised almost without errors. A crumpled receipt photographed in poor light may be half wrong. Handwriting is not recognised reliably by general-purpose OCR at all.
Getting a Better Scan
- Scan at 300 dpi, in greyscale or black and white. Higher resolutions make large files and rarely improve recognition of normal print.
- Use a phone scanning app rather than the camera. Such apps straighten the page, remove shadows and increase contrast. The built-in Notes app on iPhone and Google Drive on Android both have one.
- Photograph from directly above, in even light, with the page filling the frame.
- Flatten the paper. For books, press the page down or scan one page at a time.
- Loose photos of pages? Combine them into one file with Image to PDF first, then convert the PDF.
Always Proofread
OCR errors are plausible-looking, which makes them dangerous. The classic confusions are 0 and O, 1, l and I, 5 and S, rn and m, and lost or invented punctuation. In ordinary text a spelling checker catches most of them. In numbers nothing does: an amount, a date, an account number or a dosage can be wrong by one character and still look perfectly reasonable.
Read anything that matters against the original — figures, names, dates and clauses in particular — before you rely on the converted document. Password-protected PDFs have to be unlocked by their owner first; converters, including ours, refuse them. For help with other PDF tasks, see How to reduce PDF file size and our document format guide.